Difference between revisions of "Moodle Learning Analytics FAQ"

Jump to: navigation, search
(Will I start getting predictions as soon as I enable the model?: clarified static vs. ml models)
(Updated for Moodle 3.7)
Line 5: Line 5:
 
==Will I start getting predictions as soon as I enable the model?==
 
==Will I start getting predictions as soon as I enable the model?==
  
Static models do not require training, and will begin to deliver insights as soon as they are enabled (and the circumstances that trigger the model occur). Machine-learning based learning analytics models, such as [[Students at risk of dropping out]], must be trained on your site data before they can generate predictions. The Moodle Learning Analytics machine-learning engine needs historical data (previous courses or other activity, depending on the model), so it will need to be enabled on a production site or a copy of your production site. The model training process happens in the background once the model is enabled on your Moodle site. Once the model has been trained, you will start to receive predictions about current courses or other components. See [[Managing analytics]] for more information.
+
'''Static''' models do not require training, and will begin to deliver insights as soon as they are enabled (and the circumstances that trigger the model occur). '''Machine-learning based''' learning analytics models, such as [[Students at risk of dropping out]], must be trained on your site data before they can generate predictions. The Moodle Learning Analytics machine-learning engine needs historical data (previous courses or other activity, depending on the model), so it will need to be enabled on a production site or a copy of your production site. The model training process happens in the background once the model is enabled on your Moodle site. Once the model has been trained, you will start to receive predictions about current courses or other components. See [[Managing analytics]] for more information on how to manage notifications.
  
 
==Is there a way to export and import model data between sites?==
 
==Is there a way to export and import model data between sites?==
  
Model '''data''' can be [[Analytics#Export | exported]] from one site and placed in the [[Analytics#Models_output_directory | model data directory]] of a new site. We are working to collect model data from many sites to construct a training set we can provide with Moodle installations. Please see [https://moodle.org/course/view.php?id=17233#| How to Participate in Learning Analytics Data Collection] for more details.
+
Model '''training data''' can be [[Using analytics#Exporting_and_Importing_models | exported]] from one site and placed in the [[Analytics settings#Models_output_directory | model data directory]] of a new site. This consists of a file of calculated target and indicator values for each sample examined by the model, along with some header information. No personally identifying information is included, but one row per sample (e.g. per course enrolment and analysis interval) is included. We are working to collect model data from many sites to construct a training set we can provide with Moodle installations. Please see [https://moodle.org/course/view.php?id=17233#| How to Participate in Learning Analytics Data Collection] for more details.
  
 
==Is there a way to export and import the settings of a model between sites?==
 
==Is there a way to export and import the settings of a model between sites?==
  
This functionality is currently in development. Please see MDL-60944 Allow model configurations to be imported and exported.
+
{{New features}}Yes. Models can be created and tested on one site, and can be [[Using analytics#Exporting_and_Importing_models | exported]] with weights and imported to a new site. This data does not reference individual users or courses in any way, and can be safely shared with researchers or other sites. On the new site, a model can be re-trained after import, or the imported model weights can be tested on historical site data to evaluate the accuracy of the model on the new site. We are working to collect model configuration and weighting data from many sites to construct a pre-trained model we can provide with Moodle installations. Please see [https://moodle.org/course/view.php?id=17233#| How to Participate in Learning Analytics Data Collection] for more details.
  
 
==How can I disable Moodle Learning Analytics notifications, or restrict them to certain users, while still enabling models for testing purposes?==
 
==How can I disable Moodle Learning Analytics notifications, or restrict them to certain users, while still enabling models for testing purposes?==
  
Notifications go to users with the "analytics:listinsights" capability in the context of the prediction-- what this means for the default model is that notifications go to teachers in each course. What you can do is modify the "Teacher" role (editingteacher) to remove that capability. Currently that capability is also turned on for Managers by default. If you want to restrict the notifications to a few individuals, make a new role with just this capability, and assign that role to the individuals you wish to grant permission to in each context.
+
Notifications go to users with the [[Capabilities/moodle/analytics:listinsights|"analytics:listinsights" capability]] in the context of the prediction-- what this means for the [[Students at risk of dropping out]] model is that notifications go to teachers in each course. What you can do is modify the "Teacher" role (editingteacher) to remove that capability. Currently that capability is also turned on for Managers by default. If you want to restrict the notifications to a few individuals, make a new role with just this capability, and assign that role to the individuals you wish to grant permission to in each context.
  
 
==How accurate are the model predictions?==
 
==How accurate are the model predictions?==
  
This varies depending on the quality and quantity of site data (including how many activities are in each course and what percentage of the course is conducted online in Moodle). See [[Analytics#Evaluate| Analytics: Evaluate]] for more details on how to review model accuracy.
+
This varies depending on the quality and quantity of site data (including how many activities are in each course and what percentage of the course is conducted online in Moodle). See [[Using analytics#Review_evaluation_results| Using analytics: Review evaluation results]] for more details on how to review model accuracy.
  
 
==How can I create my own Learning Analytics models?==
 
==How can I create my own Learning Analytics models?==
  
At present, a new model must be implemented as a Moodle plugin in PHP. See the [https://docs.moodle.org/dev/Analytics_API Moodle Learning Analytics API] for details. We are working to develop tools that will allow non-programmers to create models.
+
New machine learning models can be created by using the Analytics API, by importing an exported model from another site, or by using the {{New features}}new web UI in 3.7. For more information, see [[Using analytics#Creating_and_editing_models| Using analytics: Creating and editing models]]. (Note: "static" models cannot be created using the web UI at this time.)
  
  

Revision as of 01:09, 28 June 2019


Here are some common questions about Moodle Learning Analytics. You may also find it helpful to visit the Moodle Learning Analytics Working Group site.

Will I start getting predictions as soon as I enable the model?

Static models do not require training, and will begin to deliver insights as soon as they are enabled (and the circumstances that trigger the model occur). Machine-learning based learning analytics models, such as Students at risk of dropping out, must be trained on your site data before they can generate predictions. The Moodle Learning Analytics machine-learning engine needs historical data (previous courses or other activity, depending on the model), so it will need to be enabled on a production site or a copy of your production site. The model training process happens in the background once the model is enabled on your Moodle site. Once the model has been trained, you will start to receive predictions about current courses or other components. See Managing analytics for more information on how to manage notifications.

Is there a way to export and import model data between sites?

Model training data can be exported from one site and placed in the model data directory of a new site. This consists of a file of calculated target and indicator values for each sample examined by the model, along with some header information. No personally identifying information is included, but one row per sample (e.g. per course enrolment and analysis interval) is included. We are working to collect model data from many sites to construct a training set we can provide with Moodle installations. Please see How to Participate in Learning Analytics Data Collection for more details.

Is there a way to export and import the settings of a model between sites?

New feature
in Moodle 3.7!
Yes. Models can be created and tested on one site, and can be exported with weights and imported to a new site. This data does not reference individual users or courses in any way, and can be safely shared with researchers or other sites. On the new site, a model can be re-trained after import, or the imported model weights can be tested on historical site data to evaluate the accuracy of the model on the new site. We are working to collect model configuration and weighting data from many sites to construct a pre-trained model we can provide with Moodle installations. Please see How to Participate in Learning Analytics Data Collection for more details.

How can I disable Moodle Learning Analytics notifications, or restrict them to certain users, while still enabling models for testing purposes?

Notifications go to users with the "analytics:listinsights" capability in the context of the prediction-- what this means for the Students at risk of dropping out model is that notifications go to teachers in each course. What you can do is modify the "Teacher" role (editingteacher) to remove that capability. Currently that capability is also turned on for Managers by default. If you want to restrict the notifications to a few individuals, make a new role with just this capability, and assign that role to the individuals you wish to grant permission to in each context.

How accurate are the model predictions?

This varies depending on the quality and quantity of site data (including how many activities are in each course and what percentage of the course is conducted online in Moodle). See Using analytics: Review evaluation results for more details on how to review model accuracy.

How can I create my own Learning Analytics models?

New machine learning models can be created by using the Analytics API, by importing an exported model from another site, or by using the New feature
in Moodle 3.7!
new web UI in 3.7. For more information, see Using analytics: Creating and editing models. (Note: "static" models cannot be created using the web UI at this time.)


See also